Using GIS To Analyze Physician Shortage Areas In Minnesota
Thomas J. Sandberg1,2
1
Department of Resource Analysis, Saint Mary’s University of Minnesota, Minneapolis,
MN 55410; 2Blue Cross Blue Shield of Minnesota, Eagan, MN 55122
Keywords: GIS, Healthcare, Health Plans, Demographic Analysis, Access, Medically
Underserved Area, Physician Shortage, Rural Health Care, Health Professional Shortage
Areas (HPSA), Physician Assistant, Advanced Nurse Practitioner, Primary Care.
Abstract
Physician shortages are a looming national problem. The current landscape of physicians
in Minnesota is one that varies by region and within counties. Minnesota has several
metropolitan areas that serve as bases for large provider concentrations and rural parts of
the state where provider coverage is scant. The state has enough physicians to adequately
serve the needs of its population; however, there is a problem of distribution. There is an
assumption that paraprofessionals make up for some of the physician shortages, but the
geographic extent is unknown.
areas with large racial populations such
as tribal areas.
Counties are among the most
common geographic units for measuring,
however, variation exists that makes
standardization difficult. This research
attempt will introduce a new
standardized geographic unit to base
HPSAs off of and compare these results
to the county-based measurement
system.
The relatively recent introduction
of advanced paraprofessionals
(physician assistants and advanced nurse
practitioners, primarily) as primary care
deliverers has an impact on HPSAs by
increasing the headcount of providers. It
is unknown if these paraprofessionals
have an impact on the geographic
distribution of shortage areas.
Introduction
Physician shortages exist in a spatial
context, and measuring them has been an
inconsistent task. For the purposes of
determining areas of physician shortage,
the US Department of Health and
Human Services has established Health
Professional Shortage Areas (HPSAs).
In an act of Congress, HPSAs were
originally created in the late 1960’s.
Only minimal adjustments have occurred
over the past 40 years. The earliest
record of a set ratio is from legislation in
1972 that indicated a 4,000:1 ratio (US
Congress, Office of Technology and
Assessment, 1990). Current HPSAs
were defined in 1980 and are defined as
regions (states, counties, census tracts)
where a 1:3500 ratio of primary care
physicians to population are not met.
Other non-geographic criteria exist for
measuring HPSAs that include areas
with large institutional populations
(prisons, state hospitals), areas with
unusually high population density and
Background
This research focuses only on the state
of Minnesota and its population.
Shortage areas are a national issue, with
1
Reimbursement (PA&R) database
courtesy of Blue Cross Blue Shield of
Minnesota (BCBSM). HPSAs are
calculated using data available internally
to the US Department of Health and
Human Services Bureau of Health
Professions (2005).
A non-governmental,
geographical unit of analysis is
introduced in this research. In order to
gauge physician shortages on a more
homogenous level, a vector-based grid
was created. This grid allows for an
even spatial distribution of geographic
units, unlike the irregular distribution of
counties or tracts. Cell size of the grid is
set at 30 by 30 miles and sets out to use
standards set forth by both the US
Department of Health and Human
Services. US Population, 2000 from the
census tract shapefile will be reassigned
into grid cells.
great variation from state to state.
Certain states, mostly those in the
Western half of the United States, have
great variation in the area, shape and
population of their counties.
Many issues can contribute to an
area being designated as a shortage area.
This research focuses solely on the
geographic aspect and does not seek to
answer why any specific HPSAs exist.
One specific limitation of the
federally defined, county-based system
is that it does not account for physicians
that traverse multiple states. Providers
from Wisconsin, Iowa, North Dakota,
and South Dakota that cross the
Minnesota border to practice will be
captured in the Minnesota provider data
set from Blue Cross Blue Shield of
Minnesota (BCBSM). Conversely, there
is no way to capture the extent of
Minnesota residents that cross into these
states and receive primary care services.
For the purposes of this model, it will be
assumed that health care consumption
occurs within the state border of
Minnesota.
Software Requirements and Data
Manipulation
All GIS functions included in this
research were performed using ArcGIS
9.1. The ET Vector Grid extension,
available from www.esri.com, was used
to create an arbitrary, homogenous
geographic unit of analysis. To calculate
shortages within individual grid cells,
the general population by census tract
needed to be reprojected into the grid
cells. A sequence of data preparation
was executed in order to estimate
population.
The census tract shapefile was
the source for all non-physician based
statistics including area and population.
This file in its normal state is projected
in the GCS North American 1983
latitude/longitude projection. In order to
represent Minnesota’s parallel and
perpendicular borders, particularly its
Methods
Data Acquisition
The federal HPSA model uses counties
and census tracts as its primary
geographic unit. These units of analysis
are available via a data set included from
the Environmental Systems Research
Institute’s (ESRI) ArcGIS software
platform. US Population, 2000 is an
attribute of the US Census Tract
shapefile and is used for all references to
population in this research.
The distribution of physicians
and paraprofessionals are calculated
using an October 2005 output from the
Provider Administration and
2
long, straight Southern border and to
avoid the East-West distortion inherent
in latitude/longitude projections, the
census tract shapefile was reprojected
into a Mercator projection.
Once the tract shapefile was
reprojected, the new grid was created
using the ArcGIS extension,
ETGeoWizards (figure 1).
ETGeoWizards is a free extension
available via the Environmental Systems
Research Institute’s (ESRI) website.
This is a free alternative to the Spatial
Analyst extension, available from ESRI
for a fee. Figure 1 shows the inputs
required by this extension to produce the
grid. Grid extent was calculated based
on the newly reprojected MN tract
shapefile. Additionally, the Mercator
projection uses meters as its unit of
measurement; the grid cell size field is
adjusted accordingly (48,240 meters is
approximately 30 miles). Once the cell
size, extent and file type (this example
uses polygons) was established, a vector
grid was created.
shapefile contains 234 polygons.
To most accurately attribute
population into the grid cells, this project
assumed that population was evenly
distributed within census tracts. In
reality, this is not always the case,
however, high density population centers
tend to have very small tracts based on
area and the likelihood of these small
tracts being divided among multiple grid
cells is small. It was assumed that
census tracts that were more likely to fall
across multiple grid cells are the large,
less populated tracts. The veracity of
this theory can be tested by analyzing
the number of unsplit tracts and by
analyzing the percent of population that
is wholly attributed into grid cells.
Figure 2. Projected 30x30 mile grid.
Figure 1. ETGeoWizards Vector Grid dialog
box.
As previously stated, strong
perpendicular and parallel border lines
minimizes irregular grid cell line up,
which was especially important when
the cells were intersected as described
later. One negative aspect that the
Mercator projection has on any analysis
is that it distorts area, especially over
large extents like the state of Minnesota.
For illustrative purposes, figure 2
shows the grid placement over the
Minnesota Counties shapefile. Only
cells that intersected or were contained
in the county shapefile were kept, cells
that did not intersect any counties were
deleted in edit mode. The final grid
3
named F_NAME_1. After the areas
were calculated, another new field
Area will be important when the
population is attributed from census
tracts to grid cells based on percentage
of area. To minimize the distortion of
area, the tract shapefile and grid
shapefile were temporarily reprojected
into the Albers Equal Area projection.
The Calculate Areas function of
ArcGIS was used to calculate square
meters of individual tracts in the census
tract shapefile. Areas were also
calculated for the vector grid shapefile to
check the validity of the grid cell size.
Because the grid shapefile was created in
a Mercator projection and due to the
nature of projection in general, there will
be some natural distortion of area.
However, due to the uniformity of the
grid cells, variation will be minimal. A
new column in the tract shapefile named
F_AREA was created by the Calculate
Areas function. Based on the projection,
the calculated values are in square
meters.
To calculate the percentage of a
tract that falls into each grid cell, new
polygons needed to be created that were
essentially the combination of the two
files. To accomplish this, the Intersect
function was used to create a new
shapefile that combined the grid and
tract shapefiles.
Figure 3 shows the resulting
polygon file created by the intersect
function. The combined file contained
2,266 polygons. Data derivation shows
that 60% of all tracts were unintersected.
This also accounted for 60% of the
population. Further derivation shows
that the denser the population of a tract
was, the less likely it was to be split just
as previously hypothesized.
The calculate areas function was
used to calculate areas of the tract-grid
intersect polygons of this new file and
attributes are collected in a new field
Figure 3. Tract/Grid intersection.
named PERCENT was manually created
using ArcGIS. Values for this field are
equal to the percentage of the tract that
exists in the corresponding grid cell.
The equation used to populate this field
was:
[PERCENT] = [F_AREA_1] / [F_AREA].
The resulting percentage, as seen
in figure 4, will be used to calculate the
2000 Population column to achieve
population by intersected polygon. The
next step was to then summarize the file
by adding population back into the grid
cells. The Dissolve command in ArcGIS
was used to create the new vector grid
shapefile, the old GRID_ID field being
the Dissolve By field.
For data derivation purposes, the
population is displayed as seen in figure
5. This map essentially mimics a
population density map. This map will
be helpful in identifying population
4
medicine, pediatrics and geriatrics.
The second set was limited to
physician extenders and included
physician assistants and advanced
nursing practitioners. Latitude and
longitude were assigned after being
processed by Centrus Desktop
geocoding software (Boulder, CO). 92%
of all practitioner records were matched
to the addresses and the remaining 8%
were assigned to the centroid of the zip
code available from the data set. It is
important to note that this 8% is
distributed throughout the state and not
concentrated in any particular area.
Figure 4. Example of percent of land area
recalculation into population by grid.
centers and gives the viewer an idea of
the distribution of people throughout the
state of Minnesota. For consistency and
aesthetic purposes, the new vector grid
shapefile was reprojected back into the
Mercator projection.
Figure 6. Geocoded location of physicians.
Figure 6 shows the distribution
of physicians in the state of
Minnesota.By spatially joining the
practitioner shapefile to the vector grid
cell shapefile, a sum of the number of
practitioners was obtained by grid cell.
Because this file captured all of the
known practice sites for a practitioner,
individuals may be counted more than
once. This data sought only to provide a
representation of providers by location.
Figure 5. 2000 Population by grid cell.
The final data set that needed to
be processed was the practitioner
distribution file from BCBSM. Fields
available include specialty type,
practitioner name and practice address.
The file was first limited to primary care
defined as the following specialty types:
general practice, family practice, internal
5
It was assumed that most of the over
representation of practitioners occurred
in highly populated areas, areas that
were unlikely to experience shortages.
Analysis
Using county and tract data to calculate
HPSAs can be misleading. Figure 7
shows 2005 HPSAs according to the US
Department of Health and Human
Services’ Bureau of Health Professions.
Under this scenario, 32 counties were
listed as either partial or full HPSAs.
This suggests that over 41% of the
Minnesota population was at least
partially medically understaffed for
primary care. The area that is
considered fully under staffed comprises
14.2% of the state’s land area and 3.3%
of the Minnesota population.
An analysis of the provider data
file from BCBSM showed a much
different landscape. Figure 8 shows a
distribution that is concentrated to just a
few counties. Under this model, four
counties accounted for 4% of the area
and 1.1% of the population, a 65%
percent decrease from the federally
defined, full HPSAs.
Historical figures for 1988 listed
the state of Minnesota as having 25 full
and partial Health Manpower Shortage
Areas (HMSAs – later renamed HPSAs).
Unfortunately, individual counties were
not listed. However, the total population
affected was listed. A total of 243,199
persons lived in areas designated as
either partial or full HMSAs (US
Congress, Office of Technology and
Assessment, 1990).
When compared against the 1990
US Census, this population accounted
for 5.6% of the total. For comparison,
the United States as a whole in 1988 had
31,357,445 persons living in HMSAs for
Figure 7. 2005 Health Professional Shortage
Areas.
a total of 12.6% of the population.
Recent research suggests that the current
US percent is closer to 20%, an increase
over the last decade (Laditka, et al.,
2005). Minnesota is similar to the
national trend. Although the exact
methodology could not be obtained for
calculating the partially under-staffed
population in 1988, the 3.3% of the fully
under-staffed population should be an
increase over current numbers. For
reference, historical practitioner data sets
beyond the year 2000 were not available
from BCBSM.
HPSA ratios, when applied to
counties and other governmental
boundaries, can produce skewed results.
A county with a large area, such as St.
Louis in Northeast Minnesota, qualifies
as a partial HPSA. Further analysis
showed that the population in this county
is concentrated in two areas, Duluth in
the extreme South and the Iron Range in
the center. The federal, county-based
model assumes that population is evenly
distributed throughout its 6,737 square
miles. However, anyone familiar with
6
created that populates with a “Y” any
record that does not meet the 1:3500
ratio.
Figure 9 shows the distribution
of grid cells that do not meet HPSA
standards using the BCBSM primary
care physician data set. Summary of the
cells indicated that 192,824 people or
3.9% of the Minnesota population
resided in these cells. Comparison
against the federally defined HPSAs
shows a 16.0% difference between the
two populations.
the Boundary Waters Canoe Area knows
that vast portions of this county are
uninhabited. The application of grid
cells allowed for a standardization of
area. Grid cells also attempted to mimic
user consumption patterns. The
Minnesota Department of Health, in its
regulation of managed care
organizations, suggests that primary care
be accessible within 30 miles. By
creating cells 30 miles square, the model
attempts to capture this effect.
Figure 8. Health Professional Shortage Areas
(HPSAs) using September 2005 Blue Cross Blue
Shield of Minnesota Physician Data.
Figure 9. Physician shortages by grid cell using
the BCBSM primary care physician data set.
Calculating shortage areas in grid
cells involved creating a few new fields
and populating them with simple
formulas. PHYSSHORTRT is the field
that captured the ratio for each grid cell.
The calculation was simply the
attributed population field divided by the
physician count and is expressed as:
One particular criteria for an area
to be defined as an HPSA is that
“primary care in contiguous areas are
over utilized, excessively distant or
otherwise inaccessible to the population
of the area under consideration.” (US
Congress, Office of Technology and
Assessment, 1990). Adjustment for grid
cells that were adjacent to metropolitan
areas was a process that attempted to
capture the influence of having a large
provider concentration within reasonable
distance. To determine the locations of
[PHYSSHORTRT] = [ATTPOP200] /
[PHYSCNT].
A second field, PHYSSHORTFL was
7
scenario, 129,823 people or 2.6% of the
state’s population was classified as
living in an HPSA. This is a 48.5%
decrease in population over the model
without metropolitan adjustment and a
24.8% decrease in population over the
federally defined county model.
metropolitan centers, the distribution of
cities with a population greater than
20,000 was displayed.
Figure 10 displays the manual
selection of grid cells based on
containment or immediate proximity to a
population center. This selection is
subjective, as no algorithm was
available. Simply choosing cells that
contained or bordered a population
center greater than 20,000 would have
lead to an overstatement of grid cells.
Choosing cells that only contained a
population center would have likely
understated areas. A new field named
METRO_ADJ was created that
identified these subjectively selected
cells and the results were displayed in
figure 10.
Figure 11. Physician shortages by grid cell,
adjusted for proximity to metropolitan areas.
The use of paraprofessionals as
primary care providers, most commonly
as physician assistants and advanced
nursing practitioners, has become a more
acceptable practice. Deregulation and
advanced training are allowing these
individuals to participate more fully in
the care of patients (Wessel, 2005).
The recent ability to prescribe
drugs and provide education to patients
has revolutionized this field. While their
impact in patient care can be measured,
it is relatively unknown what impact
these practitioners have on the
geographic distribution of shortage
areas. In their research, Cooper, et al.
(2002) suggest that physician extenders
are providing care in rural areas where
physician shortages are likely.
Figure 10. Cells considered “metropolitan” or
adjacent to a metropolitan cell.
Figure 11 shows the physician
distribution after the metropolitan area
adjustment. A new field was created in
ArcGIS named PHYSSHORTADJ that
essentially voided an HPSA if it also
was a cell that was metropolitan adjusted
(METRO_ADJ=Y). Under this
8
2.3% of the total population. The metro
adjustment was a 43.3% decrease from
the unadjusted combined group. When
compared to the physician only,
unadjusted model, this scenario
represents a 67.9% decrease in the
population affected.
The addition of the
paraprofessional population did in fact
change the landscape of shortage areas,
although not dramatically, as shown in
figure 12. In this scenario, only 4 cells
toggled from shortage to surplus. These
4 cells were not concentrated in any
particular area, but were distributed
throughout the state. The population
now considered to be in shortage areas
was 164,645, or approximately 3.3% of
the total population. This accounts for a
17.1% improvement over the unadjusted,
physician only model as was previously
shown in figure 9.
Figure 13. Physician/physician extender
shortages by grid cell, adjusted for proximity to
metropolitan areas.
Analytical comparison of the
federally defined full HPSAs and the
metropolitan adjusted combined group
grid cells shows a 41.1% difference in
the population affected. This suggests
that either the HPSAs over inflated
shortage areas with 162,048 people
affected or that the shortage cell model
under estimated these areas with 114,868
Minnesotan’s affected. While the
federal model is certainly easier to
administrate by using widely available,
county based data; the cell based
analysis can show more accurate
shortage distributions.
The practitioner population,
similar to the general population, is one
that is aging. Medical school output has
Figure 12. Physician/physician extender
shortages by grid cell.
When the same metropolitan
adjustment was applied to this combined
practitioner group, results were
comparable to the physician only model.
Visual analysis of figure 13, however,
shows that these shortages were more
regionalized than in previous examples.
Southern and much of Eastern
Minnesota was void of any shortage
areas. Shortage cells under this scenario
now account for 114,868 people, or
9
populated by the results of the following
equation:
remained relatively static and has not
kept pace with the growth of the general
population (Cooper, et al., 2002). Figure
14 shows the distribution of areas at risk
for impending retirements.
[AGE_FLAG] = [PhysAvAge] > 54.
The geographic distribution of these
shortages is now focused on the
Southern and Western parts of
Minnesota, contrary to the distribution of
HPSAs. The Minnesota population that
will be affected is 60,244, or 1.2% of the
total population.
Results
By combining the results of the
metropolitan adjusted HPSAs and the
impending physician retirements by grid
cells, a new pattern arises. Figure 15
shows that this distribution no longer is
focused in Northwest Minnesota, but
now includes a presence in Southern
Minnesota.
Figure 14. Areas of high risk/impending
physician retirements.
This model summarized physicians by
age to create an average by grid cell.
Rates were not adjusted by using
population projections for 2010 or
beyond and were not adjusted for
proximity to metropolitan areas. This
model exists only to predict where
shortages may occur based on
impending physician retirements. This
model also did not account for the
impact that paraprofessionals will have
on these grid cells. There is no simple
way to predict where this population
may move to and also the ages of
physician extenders and nurse
practitioners were unavailable.
Cells that have average ages
greater than 54 years or older were
flagged as high risk. A new field named
AGE_FLAG was created and was
Figure 15. Current shortages and impending
retirements.
Under this scenario, 157,493 people or
3.2% of the population lived in a current
or impending shortage area. This
10
HPSAs against grid cells, the geographic
distribution varied greatly. The use of a
standardized geographic unit not only
allowed easy visual analysis; it also
estimated density more accurately than
the county/tract based model of the
HPSA.
A certain limitation of the grid
cell model is that it would need to be
accepted at a large-scale level.
Alignment of cells, size of cells, map
projection and data source of the
population are all aspects that would
certainly introduce variation and thus
different results. This model will work
best in a controlled situation where
results can easily be repeated.
accounted for a mere 2.9% decrease
from the current, federally defined
HPSA model. While state figures remain
similar, the grid cell model was better
equipped to analyze micro scale shortage
situations.
Appendix A provides a
summarization of the statistics presented
in this analysis. Statistics were
calculated using ArcGIS 9.1 and
Microsoft Access 97.
Deliverables
Since it is county health boards that file
to declare HPSAs, it is unlikely that a
model that is not county based would be
implemented on the state or national
level. Non governmental agencies such
as health plans may find use in such a
model for a number of different
situations. This model may be useful
when looking at shortages of provider
specialties. Rather than use the general
population, membership could be
substituted. This model would be an
improvement over simple buffer models
that have a difficult time portraying
density of either members or providers.
Health plans may also find this
model useful when analyzing contracting
and marketing opportunities. Areas that
have high provider counts and low
membership may be targeted at open
enrollment time. This data combined
with employer locations and census
demographic variables could be a useful
tool for direct marketing purposes.
Acknowledgements
I would like to thank Nancy Garrett,
Boyd Lebow and Lynn Brown of Blue
Cross Blue Shield of Minnesota for their
help and advice towards the direction of
this project. I would like to also thank
Blue Cross and Blue Shield of
Minnesota for allowing me the use of
their provider data set and for supporting
me in my career and educational
development. Finally, I would like to
thank my colleagues at St. Mary’s
University of Minnesota and the staff,
especially John Ebert, Patrick Thorsell
and Dr. Dave McConville for their
project support, editing and
encouragement.
References
Cooper, R., Getzen, T., McKee, H.,
Laud, P. 2002. Economic and
Demographic Trends Signal An
Impending Physician Shortage. Health
Affairs, 21 (1): 140-154.
Laditka, J., Laditka, S., Probst, J. 2005.
More May Be Better: Evidence of a
Discussion and Conclusion
The introduction of a new geographic
unit produced very different results from
the federally defined, county/tract based
model. While the overall numbers only
changed slightly when comparing
11
Negative Relationship between
Physician Supply and Hospitalization
for Ambulatory Care Sensitive
Conditions. Health Services Research,
40(4): 1148-1166.
U.S. Congress, Office of Technology
Assessment. 1990. Health Care in
Rural America,OTA-H-434. U.S.
Government Printing Office,
Washington, DC.
U.S. Department of Health and Human
Services, Bureau of Health Professions.
2005. Health Professional Shortage
Area Primary Medical Care Designation
Criteria, 42CFR. Retrieved October,
2005 from http://bhpr.hrsa.gov/
shortage/hpsacritpcm.htm.
Wessel, L. 2005. Nurse Practitioners in
Community Health Settings Today.
Journal of Health Care for the Poor
and Underserved, 16: 1-6.
12
Appendix A. Summarization of Statistics.
Geography
State of MN
HPSA – Full County, 2005
HPSA – Partial County, 2005
HPSA – Combined, 2005
BCBSM HPSA – Full, 2005
HMSA – Combined, 1988
Grid Cells
Physician Shortage
Physician Shortage - metropolitan adjusted
Combined Shortage
Combined Shortage – metropolitan adjusted
Impending shortage (retirement)
Final shortage (combined adjusted, retirement)
Counties
87
12
20
32
4
25
Population
4,919,471
162,048
1,886,537
2,048,585
55,818
243,199
% of Total
100.0%
3.9%
38.3%
41.6%
1.1%
5.6%
Cells
24
19
20
15
10
24
192,824
129,823
164,645
114,868
60,244
157,493
3.9%
2.6%
3.3%
2.3%
1.2%
3.2%
Key Findings:
• 2005 Partial and Combined HPSAs are heavily skewed with the addition of Hennepin
County. According to the US Department of Health and Human Services website,
three extremely dense, urban census tracts meet special, non-geographic criteria that
designate this county as an HPSA. The county population is just over 1 million
persons and accounts for over 50% of the 20 partial county count. Removal of this
county changes the HPSA partial, and the HPSA combined percent of total columns
to 15.7% and 19.0%, respectively.
• There is a slight decrease in the percent of population living in HPSAs when
paraprofessionals are added.
• There is a slight decrease in the percent of population living in HPSAs when
comparing the full county, federal model to the combined, metropolitan adjusted
vector grid cell model.
• While there appears to be an improvement from the historical 5.6% HMSA count,
adjustment for impending retirements indicates that the percent of population in
HPSAs will increase in the future.
13